The dilation size of the first 3x3 convolution in DilatedResNetwork

torchcv/model/backbone/resnet/resnet_backbone.py

Lines 78 to 82 in e306e0b

    
           if m.stride == (2, 2): 
        
               m.stride = (1, 1) 
        
               if m.kernel_size == (3, 3): 
        
                   m.dilation = (dilate // 2, dilate // 2) 
        
                   m.padding = (dilate // 2, dilate // 2)

In the Dilated Resnet used for segmentation tasks, the m.dilation and m.padding of the first 3x3 convolution in layer3 and layer4 are set as half of the expected dilation ( dilation = dilation // 2).
But in the other implementations, the m.dilation and m.padding of the 3x3conv in corresponding position are set as m.dilation = dilation. Which one is as same as the original design in the Deeplab?

	if m.stride == (2, 2):
	m.stride = (1, 1)
	if m.kernel_size == (3, 3):
	m.dilation = (dilate // 2, dilate // 2)
	m.padding = (dilate // 2, dilate // 2)